#karpathy
13 articles on this topic.
-
Context Engineering Is the New Prompt Engineering
Prompt engineering quietly died in 2025. The discipline that replaced it — context engineering — is what separates a generic AI assistant from one that actually does your work. Inside Karpathy's OS analogy, Anthropic's playbook, and why the working directory is now the moat.
-
Build Your Own Wikipedia: Karpathy's Wiki LLM Explained
Karpathy's Wiki LLM concept: explicit local knowledge beats opaque AI memory. Build your own personal Wikipedia from any webpage, Apple Notes, or diary — in Markdown, for any AI.
-
Karpathy's 'Two Groups' of AI Users --- Which One Are You?
Andrej Karpathy says there's a growing gap in understanding of AI capability. One group thinks AI is a toy. The other is experiencing 'AI Psychosis.' Here's what separates them --- and how to cross the divide.
-
Markdown Wikis Are Replacing RAG — Karpathy's Pattern Explained
A folder of Markdown files beats a vector database for personal knowledge. Why Karpathy's post-RAG pattern works, when to use it, and how to build one in 15 minutes with Minibase Vault.
-
Karpathy Uses Obsidian Web Clipper. Minibase Vault Does the Same Thing — With Claude Built In
Andrej Karpathy's viral LLM Knowledge Base workflow needs a web clipper, a folder, and an MCP server. Minibase Vault is all three, in one toggle.
-
Autoresearch for Everyone: How to Run 100 AI Experiments While You Sleep
Karpathy's autoresearch runs 100+ ML experiments overnight on a single GPU. Here's how it works, what you need, and why a 630-line Python script is changing AI research.
-
The Git Commit as Scientific Discovery: How Autoresearch Turns Version Control into a Research Lab
In Karpathy's autoresearch, every successful experiment is a git commit. Every failed one is a git reset. Version control has become the memory of autonomous AI research.
-
How to Write a Good program.md: A Practical Guide for AI Agent Instructions
program.md is the file that programs AI agents in Karpathy's autoresearch. Here's how to write one that gets results --- with structure, examples, and best practices.
-
Karpathy's Autoresearch & PROGRAM.md: AI That Runs Experiments While You Sleep
Andrej Karpathy's autoresearch lets AI agents run 100+ ML experiments overnight, guided by a single Markdown file called program.md. Here's how it works and why it matters.
-
From README.md to PROGRAM.md: Markdown Is Now a Programming Language
README.md was for humans. AGENTS.md is for coding assistants. PROGRAM.md is for autonomous research. Markdown has evolved from documentation into the programming language for AI agents.
-
From SETI@home to AgentHub: Karpathy's Vision for Distributed AI Research
Karpathy says the next step for autoresearch is SETI@home-style distributed collaboration. AgentHub and Hyperspace are already making it real. Here's what it means.
-
Shopify's 19% Improvement: How Companies Are Already Using the Autoresearch Pattern
Days after Karpathy released autoresearch, Shopify's CEO ran it on their own models and saw a 19% improvement. Here's how companies are adopting autonomous AI experimentation.
-
Vibe Coding Is Dead. Welcome to Agentic Engineering with Markdown.
Karpathy says vibe coding is passé. The new paradigm is agentic engineering, where you orchestrate AI agents through Markdown files instead of writing code yourself.